The increasing sophistication of cyber threats from AI-driven adversarial attacks to quantum-enabled exploits has revealed critical limitations in conventional network anomaly detection (NAD) and intrusion detection systems (IDS). This review addresses a gap in existing literature through its synthesis of advancements from 2015 to 2024. It systematically evaluates the interplay between technological innovation, evolving attack vectors, and also regulatory constraints. Our analysis, unlike prior surveys, covers methodological evolution, ethical-compliance challenges, operational scalability, and emerging threat landscapes. By cataloging over 120 peer-reviewed studies, alongside industry reports, we identify further model shifts to federated learning in decentralized threat analysis, also graph neural networks (GNNs) to track advanced persistent threats (APTs), with homomorphic encryption in real-time inspection regarding encrypted traffic. Enduring barriers involve biases in ML training datasets, interoperability gaps inside hybrid systems, as well as the absence of standardized benchmarks for AI-driven IDS. 

The review critiques the disconnect that is between academic research and industrial deployment, supporting lightweight and explainable models for resource-constrained networks. We propose one taxonomy of next-generation NAD/IDS architectures stressing zero-trust principles, adversarial resilience, and human-in-the-loop validation. The work underscores the urgency of international collaboration to establish open threat intelligence repositories. It also highlights regulatory sandboxes, ensuring cybersecurity innovation aligns with global imperatives.